基于稀疏表示的声频目标识别方法研究  

Research of acoustic target recognition method based on sparse representation

作　　者：刁志蕙 冯玉田[1] 王瑞[1] 毕超[1] 李园辉[1] 

机构地区：[1]上海大学通信与信息工程学院,上海200444

出　　处：《电声技术》2016年第5期31-34,共4页Audio Engineering

摘　　要：提出分别利用短时傅里叶变换和小波变换进行特征提取和稀疏表示分类(SRC)的车辆识别方法。其中,短时傅里叶变换(STFT)和离散小波变换(DWT)分别从每个传感节点收集到的声音信息中提取车辆的特征向量,SRC通过特征训练集建立一个过完备字典来求解稀疏最优化问题,从而实现分类识别。实验结果表明,短时傅里叶变换提取特征并进行分类的效果高于用小波变换进行特征提取并分类的方法,也高于利用MFCC提取车辆声音特征并进行分类的方法。A vehicle recognition method is proposed based on feature extraction of Short-Time Fourier transform（STFT） and Discrete Wavelet Transform（DWT） and Sparse Representation Classification （SRC） in acoustic sensor networks. In the method, STFT and DWT are used to extract the vehicle feature vectors from the acoustic information collected by individual sensor nodes.Then, SRC is used to accomplish classification throughtraining feature sets and establishing an over-complete dictionary.Experimental results show that the STFT feature extraction and SRC classification approach has higher recognition rate compared with MFCC and DWT methods.

关 键 词：短时傅里叶变换 离散小波变换 稀疏表示 过完备字典 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]